1. Individual Particle Reconstruction: PFA Development in the US Norman Graf (for M. Charles, L. Xia, S. Magill) LCWS07 June 2, 2007

2. 2 Overview PFA reconstruction is complex & you have to get many individual steps right: track finding, fitting & extrapolation track-cluster matching MIP identification photon identification hadronic shower clustering handling of displaced secondaries calibration of photons calibration of neutral hadrons E/p cut (including calibration) PFA development isn't about finding the "magic bullet" perfect algorithm. It's about iteratively finding the worst problems that are limiting performance fixing them hopefully seeing things improve a little finding the next worst problems Interplay between detector and algorithm makes it tricky to really tune detector design you need a really good (mistake-free) algorithm fair comparison means equal tuning on different detectors But we can say: you can do at least this well with this detector.

3. 3 Analysis Tools Common input data samples  Single particles for tuning detector response  Dijet samples (uds) @ 91, 200, 500 GeV cms  e + e -  ZZ  ( n) ( q q) Common detector simulations Provide a number of QA tools to assure that some common tasks are handled in the same way. Calorimeter Calibrations  Sampling fractions & common energy corrections Perfect PFA performance  Common definitions of “final state” particles  Based on Generator or Simulated Particles?  Standard cheated tracks, cheated clusters

4. 4 Analyses Presenting three different analyses  Mat Charles (U. Iowa)  Lei Xia (ANL)  Steve Magill (ANL), Norman Graf (SLAC) Detectors are variants of SiD concept.  But framework supports essentially arbitrary detectors and plan is to explore larger phase space. More details in slides presented at recent SiD workshop at FNAL.

5. 5 Algorithm Description I Main philosophy is to tackle the (relatively) easy problems first. Step 1: Find photons, remove their hits. Tight clustering Apply shower size, shape, position cuts (very soft photons fail these) Make sure that they aren’t connected to a charged track Step 2: Identify MIPs/track segments in calorimeters. Identify dense clumps of hits. These are the building blocks for hadronic showers Pretty easy to define & find Step 3: Reconstruct skeleton hadronic showers Coarse clustering to find shower components (track segments, clumps) that are nearby Use geometrical information in likelihood selector to see if pairs of components are connected Build topologically connected skeletons If >1 track connected to a skeleton, go back and cut links to separate Muons and electrons implicitly included in this step too Step 4: Flesh out showers with nearby hits Proximity-based clustering with 3cm threshold Step 5: Identify charged primaries, neutral primaries, soft photons, fragments Extrapolate tracks to clusters to find charged primaries Look at size, pointing, position to discriminate between other cases Merge fragments into nearest primary Use E/p veto on track-cluster matching to reject mistakes (inefficient but mostly unbiased) Use calibration to get mass for neutrals & for charged clusters without a track match (calibrations for EM, hadronic showers provided by Ron Cassell) Known issues & planned improvements: Still some cases when multiple tracks get assigned to a single cluster Punch-through (muons and energetic/late-showering hadrons) confuses E/p cut Improve photon reconstruction & ID Improve shower likelihood (more geometry input) Use real tracking when available No real charged PID done at this point M. Charles

6. 6 Current Performance Looking at: e + e -  Z1 ( n) Z2 ( q q) for q=u,d,s at √ s=500 GeV requiring primary quarks have |cos(theta)|<0.8 reconstructing dijet invariant mass, i.e. mass of Z2 quoting residual = (true mass of Z2 - reconstructed mass of Z2) M. Charles

7. 7 SiD SS/Scin HCAL SiD W/Scin HCALSiD W/RPC HCAL SiD SS/RPC HCAL M. Charles

8. 8 Current Performance Little discriminations between designs at this point Working to improve performance… … but not actually too far from “glass ceiling” of 4.1 GeV (for W/RPC). To understand/approach/move beyond that, need to think more broadly:  Is assumed tracking performance realistic? Too pessimistic? Too optimistic?  Can calibration be improved?  Can we put in more information? E.g. pick up low-pt tracks that are being ignored  Is this event type (with boosted jets) representative?  What physics models are being used for the showers? (lcphys vs lhep vs…) rms90mean90 W/Scint5.4 GeV-3.5 GeV W/RPC5.4 GeV-3.1 GeV SS/Scint5.2 GeV-2.2 GeV SS/RPC5.4 GeV-1.7 GeV M. Charles

9. 9 Performance Caveats Numbers depend on things that are not really algorithm specific: detector physics event type Assumptions about tracking & track extrapolation polar angle & acceptance calibration how physics quantity (e.g. dijet mass) is measured how figure of merit (e.g. rms90) is computed Some things lower the ceiling rapidly. For example, with e + e -  ZZ  ( n)( q q) in |cos(theta)|<0.8 for W/RPC SiD detector, dijet mass resolution (rms90) is: 0.0 GeV if completely cheating 0.5 GeV dropping missed particles (using GenFinalStateParticles) 1.2 GeV dropping missed particles (using SimFinalStateParticles with cuts) 2.9 GeV including resolution of neutrals using Ron's Z-pole calibration 3.2 GeV requiring a track for charged particles (else treated as neutral hadrons) 3.4 GeV requiring that tracks can be extrapolated to ECAL surface 4.1 GeV requiring cluster be within 25mm (depending how track extrapolation is done) 5.1 GeV if using naive helical track extrapolation instead So different assumptions about tracking, calibration, etc can have a big impact. … unless completely confusion-dominated M. Charles

10. Algorithm description II Reconstructed TracksCalorimeter Clusters Clustering Algorithm Photon Identification EM ClustersHadron Clusters ‘ Neutral ’ ClustersMatched Clusters Track-cluster matching Charge fragment identification Neutral ClustersFragments E photon E neu-had 00P track Hadron sampling fraction EM sampling fraction Total event energy Track finding Algorithm (use MC truth for now) E/p check Calorimeter HitsTracker Hits E corr L. Xia

11. Current PFA Z-pole performance All events, no cut Mean 88.43 GeV RMS 5.718 GeV RMS90 3.600 GeV [38.2 %/sqrt(E)] Barrel events (cos(theta[Q]) < 1/sqrt(2)) Mean 89.10 GeV RMS 4.646 GeV RMS90 3.283 GeV [34.7 %/sqrt(E)] L. Xia

12. Progress on PFA performance at Z-pole improved/tuned algorithms Clustering algorithm Track extrapolation Track/cluster matching Photon finder E/P correction Fragment identification Still need improvement Clustering algorithm Photon finder Fragment attachment …(more to come) Jet Energy Resolution at Z-pole %/sqrt(E) 30 35 40 45 July Aug Sept Oct Nov Dec Jan Feb Mar Apr May June 2006 2007 L. Xia

13. Using Z-pole tuned PFA at higher energies Barrel event (cos(theta[Q]) < 1/sqrt(2)) Mean209.3 GeV RMS15.6 GeV RMS909.12 GeV [62.6%/sqrt(E)] Barrel event (cos(theta[Q]) < 1/sqrt(2)) Mean485.6 GeV RMS43.7 GeV RMS9027.6 GeV [124.%/sqrt(E)] 200 GeV 500 GeV Not good yet – but algorithms not tuned at these energy A lot of improvement expected, clearly still a lot of work to be done! L. Xia

14. Shower leakage: di-jet at 200 GeV RMS = 15.89 GeV RMS90 = 9.632 GeV [66.7%/sqrt(E)] RMS = 11.44 GeV RMS90 = 8.45 GeV [~59%/sqrt(E)] Removing events with shower leakage L. Xia

15. Shower leakage: di-jet at 500 GeV RMS = 30.25 GeV RMS90 = 21.4 GeV [~97%/sqrt(E)] RMS = 43.88 GeV RMS90 = 28.11 GeV [127.%/sqrt(E)] Removing events with shower leakage Shower leakage affect PFA performance at high energy Events with heavy shower leakage could be identified by hits in the muon detectors Use hits in the muon detectors to estimate shower leakage? L. Xia

16. Re-writing PFA according to lcsim template Motivation Facilitate exchange with other PFA efforts Check my algorithm from head to toe Write intermediate lcio output file to save running time on the rest of the PFA (do not repeat clustering for each run) Current status Program re-writing is done Algorithm is fully modular Followed lcsim template convention as closely as I can However, used some extensions of standard interface Some issues exist Z-pole result is still different from old algorithm, but the difference is very small now Some problems with the intermediate lcio file L. Xia

17. Re-writing PFA according to lcsim template Current status (continue) Program performance Overall running time is actually longer 10k Z-pole events: 10hr => 14hr If intermediate lcio successful, can save ~90% running time (by not repeating clustering each time) Will upload to lcsim cvs, after solving some obvious issues L. Xia

18. Future plans Finish up PFA tuning at Z-pole Concentrate on performance improvement at higher energies Shower leakage study using detector models with extended HCal Detector performance study with fully developed PFA L. Xia

19. 19 Algorithm Description III Track-linked mip segments (ANL)  find mip hits on extrapolated tracks, determine layer of first interaction based solely on cell density (no clustering of hits) (   candidates) Photon Finder (SLAC)  use analytic longitudinal H-matrix fit to layer E profile with ECAL clusters as input (  ,  0, e +/- candidates) Track-linked EM and HAD clusters (ANL, SLAC)  substitute for Cal objects (mips + non-EM ECAL shower clusters + HCAL calorimeter hits (or clusters) )  reconstruct linked mip segments + clusters iterated in E/p  Analog or digital techniques in HCAL (   +/- candidates) Neutral Finder algorithm (SLAC, ANL)  cluster remaining CAL cells, merge, cut fragments (  n, K 0 L candidates) Jet algorithm  Reconstructed Particles used as input to jet algorithm, further analysis S. Magill

20. 20 Track-Linked mip segments Interaction Length of Tracks # of Mip Hits per Track S. Magill

21. 21 Photon Finding  Number  Energy S. Magill

22. 22 Photon Finding Photon Energy Sum @ Z Pole S. Magill

23. 23 Neutral Hadrons Neutral Hadron Energy Sum @ Z Pole S. Magill

24. 24 Reconstruction Framework Analyses shown here done within the general ALCPG simulation & reconstruction environment. Framework exists for the full reconstruction chain which allows modular implementation of most aspects of the analysis. Interfaces allow different clustering algorithms to be swapped in and alternate strategies to be studied. Goal is to facilitate cooperative development and reduce time & effort between having an idea and seeing the results.

25. 25 Summary Individual Particle Reconstruction algorithms being developed with minimal coupling to specific detector designs.  Will allow full phase space of detector designs to be studied in a common framework. Finishing development of common infrastructure tools  Calibration method for detector models  Perfect PFA prescription Released Reconstruction Template  Enables e.g. Cluster algorithm substitution, CAL hit/cluster accounting  Migrating individual analyses into this framework Optimization & Standardization of reconstructors  Photon & muon finders fairly mature, close to release Analysis emphasis on dijet invariant mass resolution in physics events  Currently e + e -  ZZ  ( n) ( q q) (No jet combinatorics, uds) (2)  Results soon from e + e -  ZZ  ( q q) ( q q) & e + e -  ZZ n, WW n (4)   t t (6)   t t h (8) Plan to release “canned” physics analyses to reduce systematic uncertainties in e.g. jet-finding, combinatorics, constrained fits, … Closing in on detector optimization using results.

26. 26 Additional Information lcsim.org - http://www.lcsim.orghttp://www.lcsim.org ILC Forum - http://forum.linearcollider.orghttp://forum.linearcollider.org Wiki - http://confluence.slac.stanford.edu/display/ilc/Homehttp://confluence.slac.stanford.edu/display/ilc/Home org.lcsim - http://www.lcsim.org/software/lcsimhttp://www.lcsim.org/software/lcsim Software Index - http://www.lcsim.org/softwarehttp://www.lcsim.org/software Detectors - http://www.lcsim.org/detectorshttp://www.lcsim.org/detectors LCIO - http://lcio.desy.dehttp://lcio.desy.de SLIC - http://www.lcsim.org/software/slichttp://www.lcsim.org/software/slic LCDD - http://www.lcsim.org/software/lcddhttp://www.lcsim.org/software/lcdd JAS3 - http://jas.freehep.org/jas3http://jas.freehep.org/jas3 AIDA - http://aida.freehep.orghttp://aida.freehep.org WIRED - http://wired.freehep.orghttp://wired.freehep.org